MCP logo

Agent Orchestration

MCP

Model Context Protocol for exposing tools to live speech agents over a shared tool surface.

MCPToolsGemini LiveVoice agentsFunction calling
Documentation
Where it fits

MCP in your voice stack.

  1. TelephonyPhone networkInbound call
  2. AgentDuetCall + audioAnswer, stream
  3. Speech-to-speechLive modelListens, call functions, replies
  4. ToolsMCPStructured tool calls
Overview

What it contributes to the stack.

The Model Context Protocol (MCP) exposes tools that models can call during a conversation. With AgentDuet, the phone call and audio stay on AgentDuet; MCP provides the tool surface the model uses for lookups, updates, and policy-gated actions.

For setup, credentials, and API details, use the Documentation link above.

Capabilities

The technical characteristics that matter when deciding whether this provider belongs in your application.

MCP tool surface

Exposes authenticated lookups, account actions, and escalation as structured tools.

Model function calling

Tool schemas from MCP become function declarations for the live speech model.

Swap the server

Change the MCP server to change agent capabilities without rewiring call audio.

Policy guardrails

Caps and approvals stay in the tool layer so the model cannot invent unauthorized outcomes.

Common use cases

Practical scenarios where the provider's role is clear and the surrounding systems remain under application control.

Account servicing desk

Verify the caller, explain fees, apply a policy-limited waiver, or escalate.

Tool-backed voice agents

Keep business operations in MCP while the speech model handles the conversation.

Shared tool surface

Reuse the same MCP server from phone and other clients that speak MCP.

Build a voice agent.

AgentDuet handles the phone and messaging boundary. Your model runs the conversation.

Explore the SDK